THE CHU CONSTRUCTION: HISTORY OF AN IDEA MICHAEL BARR

advertisement
Theory and Applications of Categories, Vol. 17, No. 1, 2006, pp. 10–16.
THE CHU CONSTRUCTION: HISTORY OF AN IDEA
MICHAEL BARR
Abstract. This paper describes the historical background and motivation involved
in the discovery (or invention) of Chu categories.
In 1975, I began a sabbatical leave at the ETH in Zürich, with the idea of studying
duality in categories in some depth. By this, I meant not such things as the duality
between Boolean algebras and Stone spaces, nor between compact and discrete abelian
groups, but rather self-dual categories such as complete semi-lattices, finite abelian groups,
and locally compact abelian groups. Moreover, I was interested in the possibilities of
having a category that was not only self dual but one that had an internal hom and for
which the duality was implemented as the internal hom into a “dualizing object”. This
was already true for the complete semi-lattices, but not for finite abelian groups or locally
compact abelian groups. The category of finite abelian groups has an internal hom, but
lacks a dualizing object, while locally compact groups have a dualizing object, but not an
internal hom that is defined everywhere. Although you could define an abelian group of
continuous homomorphisms between locally compact abelian groups, there was no way of
systematically putting a locally compact topology on the hom set that would lead to the
desired properties.
The desired properties were what I subsequently called ∗-autonomy. I do not want to
go into the technical details here (see [Barr, 1999], for example). Basically, a ∗-autonomous
category has an internal hom, denoted −◦, a symmetric monoidal structure, ⊗, along with
the usual coherent isomorphisms Hom(A ⊗ B, C) ' Hom(A, B −◦ C). In addition, there
should be a “dualizing object”, ⊥, such that if we define A∗ = A −◦ ⊥, the subsequent
/ (B ∗ −◦ A∗ ) should be an isomorphism.
canonical map (A −◦ B)
There is a certain redundancy in this definition. For instance, the most efficient (and
perhaps the most natural) way is to assume just −◦ and ⊥ and define A∗ = A −◦ ⊥ and
then A ⊗ B = (A −◦ B ∗ )∗ , since the latter two are provably naturally isomorphic.
By the end of the year, I had in fact produced a moderate number of examples of ∗autonomous categories. One of them was a full subcategory of topological abelian groups
that included all the locally compact abelian (LCA) groups in such a way that the duality
restricted to them was the well-known duality of LCA groups. As one would expect, the
the circle group was the dualizing object. In addition, the category was complete and
cocomplete. In fact, it was built, essentially, by completing the LCA groups.
This work was supported by a grant from the NSERC of Canada
Received by the editors 2003-06-06 and, in revised form, 2006-11-09.
Published on 2006-12-16 in the volume Chu spaces: theory and applications.
2000 Mathematics Subject Classification: 18-03, 01A65, 18D15.
Key words and phrases: Chu categories, *-autonomous categories, history.
c Michael Barr, 2006. Permission to copy for private use granted.
10
THE CHU CONSTRUCTION: HISTORY OF AN IDEA
11
Another example was a full subcategory of the category of locally convex topological
vector spaces. In the process of studying that category, I read a number of books on that
subject, for example [Schaefer, 1970] and [Pietsch, 1972]. Each one contained a certain
construction of “pairs”. A pair (E, E 0 ) consists of two (discrete) vector spaces equipped
/ C, the complex numbers. This device was
with a bilinear map1 h−, −i : E × E 0
used because of the large number of possible topologies on the dual space of a topological
vector space. Basically, you were to think of E and E 0 as being dual to each other with
certain topologies.
As I was preparing an early draft of the subsequent monograph [Barr, 1979], I began to
be concerned about the fact that I had constructed this elaborate theory and had only a
bare handful of examples. Aside from the complete semi-lattices the other examples were
constructed with some difficulty using quite ad hoc methods. I had noticed the pairs and
wondered if there were a way of defining an internal hom among pairs. It turned out to
be surprisingly easy. I began with the obvious interpretation that in a pair E = (E, E 0 ),
E was the set of points and E 0 the set of functionals. (Of course, in the dual pair E 0 is
the set of points and E the set of functionals.) So if F = (F, F 0 ) is another such object,
the points of E −◦ F should certainly be the set of maps between the pairs and a map was
/ F such that for all φ ∈ F 0 , φ.f ∈ E 0 . Call this set
a continuous linear map f : E
[E, F]. Then we wanted
E −◦ F = ([E, F], ?)
(∗)
with ? an unknown to be determined. A similar guess was that the tensor product of two
objects might be given by
(E, E 0 ) ⊗ (F, F 0 ) = (E ⊗ F, ??)
(∗∗)
where ?? is another unknown to be determined. But if we put these together with the
fact that in any ∗-autonomous category,
E ⊗ F = (E −◦ F∗ )∗
(or, equivalently, that E −◦ F = (E ⊗ F∗ )∗ ), we can compute
E ⊗ F = (E −◦ F∗ )∗
= ((E, E 0 ) −◦ (F 0 , F ))∗
= ([(E, E 0 ), (F 0 , F )], ?)∗
= (?, [(E, E 0 ), (F 0 , F )])
Comparing this with (∗∗) we see that
E ⊗ F = (E ⊗ F, [(E, E 0 ), (F 0 , F )])
1
It was usually supposed that the pairing was non-singular. As a first approximation, I had to omit
that hypothesis. Later, a way was found to reinstate it, if desired, at least in the familiar examples.
12
MICHAEL BARR
Similarly, we find that
E −◦ F = ([E, F], E ⊗ F 0 )
These formulas are suggestive, but they don’t prove anything. But, amazingly, it all
works!
Thus I ended up with a category whose objects were pairs E = (E, E 0 ) of vector spaces
/ C. A map from E to F = (F, F 0 ) is a pair of linear
equipped with a pairing E ⊗ E 0
/ F and f 0 : F 0
/ E 0 (note the direction reversal) such
maps (f, f 0 ) in which f : E
0
that hf v, wi = hv, f wi whenever v ∈ E and w ∈ F 0 . There is no topology assumed and
no continuity on the linear maps.
At this point, it seemed clear that no special properties of vector spaces was being
used and this construction was an instance of a general one. Let V be an autonomous
category, meaning one which had a coherently associative, unitary, symmetric tensor
product, − ⊗ −, an internal hom − −◦ − with the usual adjunction
Hom(A ⊗ B, C) ' Hom(A, B −◦ C)
The only additional assumption we need to make on V is that it have pullbacks. Fix an
object of V , which we will call ⊥. Form the category of pairs E = (E, E 0 ) equipped with
/F
/ ⊥. If F = (F, F 0 ) is another object, then a morphism E
a morphism E ⊗ E 0
0
0
0
0
/
/
F and f : F
E such that the diagram
is a pair (f, f ) where f : E
E ⊗F0
f ⊗F 0
E⊗f 0
/ F ⊗F0
h−,−i
E ⊗ E0
h−,−i
/⊥
commutes. This condition is just a diagrammatic interpretation of the definition of morphism. It is equivalent to the assertion that the following square is a pullback:
Hom(E, F)
/ Hom(E, F )
/ Hom(E ⊗ F 0 , ⊥)
Hom(F 0 , E 0 )
This definition gives a category that is called Chu(V , ⊥). Assuming that V has pullbacks,
we can internalize the above definition, by defining [E, F] so that
[E, F]
/ E −◦ F
/ E ⊗ F 0 −◦ ⊥
F 0 −◦ E 0
13
THE CHU CONSTRUCTION: HISTORY OF AN IDEA
is a pullback. Then define
E −◦ F = ([E, F], E ⊗ F 0 )
The pairing is either of the two equal maps
[E, F] ⊗ E ⊗ F 0
/ (E −◦ F ) ⊗ E ⊗ F 0
/F ⊗F0
/⊥
or
[E, F] ⊗ E ⊗ F 0
/ (F 0 −◦ E 0 ) ⊗ E ⊗ F 0
/ E0 ⊗ E
/⊥
It seemed clear that this gave a ∗-autonomous category, but there were a number
of unpleasant details to be verified. Since my student, Po-Hsaing Chu needed a master’s
project, so I asked him to verify them, which he did [Chu, 1978, 1979]. I now had expanded
from six to infinity the repertory of known ∗-autonomous categories. At this point, the
construction, and in fact the whole idea of ∗-autonomous categories, was more-or-less
forgotten, not least by me.
One interesting thing is that I never looked to see if there was any connection between
the ∗-autonomous categories that I had carefully constructed and related ∗-autonomous
categories. For example, I had constructed a ∗-autonomous category that was a full subcategory of topological abelian groups and made no attempt to relate it to Chu(Ab, K),
where K is the circle group. This was a mistake.
Eight years later, in 1987, I became aware of Jean-Yves Girard’s linear logic. I do
not recall how I became aware of it, but at the meeting at The University of Colorado
that summer, I knew enough to bring Girard a copy of [Barr, 1979]. Since ∗-autonomous
categories give models of linear logic and since the Chu construction gives easy examples
of ∗-autonomous categories, this was reasonable, but I must have known that before going
to Boulder. I think Robert Seely must have lectured on this in our seminar that spring.
At any rate, knowledge of ∗-autonomous categories in general and the Chu construction
in particular had penetrated the linear logic and theoretical computer science community.
1. Further work
Vaughan Pratt is convinced that Chu categories are an important adjunct to the study of
a number of questions in theoretical computer science. He has written a series of papers
expanding on this perception. A small selection of them is [Pratt, 1993a, 1993b, 1995].
In the meantime, I became interested in the connection between the categories constructed in my 1979 monograph and the Chu construction. In that paper, I had constructed a small number of ∗-autonomous categories. In all but one case I had begun
with an incomplete, but ∗-autonomous category that was embedded as a full subcategory
of a complete autonomous category that lacked a duality. I then showed how there was
a kind of completion of the ∗-autonomous subcategory inside the category of topological
(or uniform space) objects of the complete autonomous category that was ∗-autonomous.
The construction was fairly difficult and it was not easy to understand which objects are
14
MICHAEL BARR
in the subcategory. Eventually, it occurred to me that if V is the category of discrete objects and ⊥ is the dualizing object, then Chu(V , ⊥) is a candidate for the ∗-autonomous
category that I had constructed.
For example, in the category of abelian groups, there is a full ∗-autonomous subcategory of the category of topological abelian groups that includes all the locally compact
groups. Let chu(Ab, K) denote the full subcategory of Chu(Ab, K) consisting of those
pairs (A, A0 ) for which the pairing is non-singular: for any a ∈ A, if a 6= 0, then there
is an a0 ∈ A0 for which ha, a0 i 6= 0 and vice versa. The definitions of the internal hom
and tensor have to be modified, but the modifications are essentially obvious. It is now
known that chu(Ab, K) is equivalent to a full subcategory of topological abelian groups
that includes all the locally compact group. Is this the same category as constructed “by
hand”? I don’t know and I have no idea how to even approach the question (and, to be
sure, there is not any reason to know the answer).
2. Mackey
I had imagined that the source of the idea of pairs of vector spaces was Grothendieck. It
looked like the kind of construction he would have done, although he might have made
a category of it. But in fact, it goes back further than that. The source seems to have
been George Mackey’s thesis, published as [Mackey, 1945]. In it he defines pairs as above,
except that his pairs are not assumed to be non-singular in the second variable. So the
dual of (E, F ) is (F, E/E0 ) where E0 consists of those vectors that are annihilated of
every functional in F .
One interesting point about Mackey’s paper is that it looks as if he was proposing
that the set of functionals be considered as a replacement for the topology, not just an
adjunct to it. After all, the topology on a set is a equivalent to a set of functions
S from the
set into the Sierpinski space 2. Not any set of functions will do; it must be a -∩-closed
sublattice of the set of all functions. In the case of the vector spaces, the condition is even
simpler; it must be a linear subspace of the space of all functionals. Note that I am not
suggesting that this is the same thing as a topological vector space. It is not, but it might
be a subject of more interest, properly studied on its own, than topological vector spaces.
At all events, if this was Mackey’s intention, it has not been followed up2 . Instead, the
various authors have used these pairs as a tool for the study of duality, rather than as
the objects of study. And the dual is, in turn, used as a tool in the study of the deeper
properties of these spaces.
3. A rational—but incorrect—reconstruction
Here is how the Chu construction should have been discovered. That it wasn’t just shows
how sometimes mathematics is two steps forward, then one back.
2
The referee informs me that some people have and mentioned books by Frölicher and Kriegl, and of
Kriegl and Michor. Interestingly Frölicher heard me lecture in this topic and didn’t mention the book.
THE CHU CONSTRUCTION: HISTORY OF AN IDEA
15
Suppose you want to make a self-dual category that contains a given category C .
Assuming that C has a terminal object 1, the category C e = C × C op is obviously
self-dual and embeds C as the objects of the form (C, 1). If C should happen to be
(symmetric) closed monoidal, then you might not expect C e to be a closed monoidal
category, but it is, at least when C has finite products. The reason you might not expect
it is that C op is not generally closed monoidal when C is. For instance, the category of
finite dimensional coalgebras over a field is closed monoidal (in fact, cartesian closed), but
the opposite category of algebras certainly is not. What is true (and it is crucial) is that
when C is closed monoidal, C op is enriched over C .
We begin with
Hom((C, D), (C 0 , D0 )) = Hom(C, C 0 ) × Hom(D0 , D)
which is the definition of the homfunctor in C e . This leads one to try
(C, D) −◦ (C 0 , D0 ) = ((C −◦ C 0 ) × (D0 −◦ D), X)
(∗)
where X = X(C, D, C 0 , D0 ) is to be determined. Similarly, you might guess that
(C, D) ⊗ (C 0 , D0 ) = (C ⊗ C 0 , Y )
where Y = Y (C, D, C 0 , D0 ) is to be determined. Naturally, you want more than just self
dual and closed monoidal, you want ∗-autonomous, which implies that
(C, D) −◦ (C 0 , D0 ) ' ((C, D) ⊗ (C 0 , D0 )∗ )∗ = ((C, D) ⊗ (D0 , C 0 ))∗
= ((C, D) ⊗ (D0 , C 0 ), Y (C, D, D0 , C 0 ))∗ = (Y (C, D, D0 , C 0 ), C ⊗ D0 )
Comparing this to (∗), we see that X(C, D, C 0 , D0 ) = C ⊗ D0 and Y (C, D, D0 , C 0 ) =
(C −◦ C 0 ) × (D0 −◦ D). This leads directly to the formulas
(C, D) −◦ (C 0 , D0 ) = ((C −◦ C 0 ) × (D0 −◦ D), C ⊗ D0 )
and
(C, D) ⊗ (C 0 , D0 ) = (C ⊗ C 0 , (C −◦ D0 ) × (C 0 −◦ D))
Amazingly, this all works! In fact, the reader might have noticed that it is just Chu(C , 1);
/ 1,
an object of the latter category is a pair (C, D), together with a morphism C ⊗ D
of which there is exactly one. This category could thus have been discovered with no
motivation from Mackey’s pairs. Could have been but wasn’t.
Moreover, this leads directly to the Chu construction. If M is a commutative monoid
object in any ∗-autonomous category, an M -action is an object A, together with an
/ A that satisfying the usual associative and unitary identities. It is
morphism M ⊗ A
not hard to prove that the category of M -actions is again a ∗-autonomous category. Any
object ⊥ of C gives a monoid object M = (>, ⊥) of C e . Since the unit for the tensor
16
MICHAEL BARR
/ 1 is the unique
product of C e is (>, 1) and the unit map for M is (id, !), where ! : ⊥
arrow. The multiplication is
(>, ⊥) ⊗ (>, ⊥) = (> ⊗ >, (> −◦ ⊥) × (> −◦ ⊥)) = (>, ⊥ × ⊥)
(id,∆)
/ (>, ⊥)
It then turns out that the category of M -actions is precisely Chu(C , ⊥).
Moreover, this constructions works with certain changes even when ⊗ is not symmetric.
In that case, two internal homs are needed in C , one to be right adjoint to C ⊗ − and the
other to be right adjoint to − ⊗ C. In that case, M need not be commutative (in fact,
the concept is undefinable) and instead of left actions of M , we must use the category
of two-sided actions. This is studied in detail in [Barr, 1995], where the ∗-autonomous
structure of C e was first described. The simplification that this construction offers is more
important in the asymmetric case than in the symmetric one.
References
M. Barr (1979), ∗-Autonomous categories. Lecture Notes in Mathematics 752 Springer-Verlag, Berlin,
Heidelberg, New York.
M. Barr (1995), Non-symmetric ∗-autonomous categories. Theoretical Computer Science, 139, 115-130.
M. Barr (1996), The Chu construction. Theory and Applications of Categories, 2, 17–35.
M. Barr (1999), ∗-Autonomous categories: once more around the track. Theory and Applications of
Categories, 6, 5–24.
M. Barr (2000), On ∗-autonomous categories of topological vector spaces. Cahiers de Topologie et Géométrie Différentielle Catégorique, 39 243–254.
P.-H. Chu (1978), Constructing ∗-autonomous categories. M. Sc. thesis, McGill University.
P.-H. Chu (1979), Constructing ∗-autonomous categories. Appendix to [Barr, 1979].
G. Mackey (1945), On infinite dimensional vector spaces. Trans. Amer. Math. Soc. 57, 155–207.
A. Pietsch, Nuclear Locally Convex Spaces. Springer-Verlag, 1972.
V.R. Pratt (1993a), Linear Logic for Generalized Quantum Mechanics. In Proceedings Workshop on
Physics and Computation, Dallas, IEEE Computer Society.
V.R. Pratt (1993b), The Second Calculus of Binary Relations. In Proceedings of MFCS’93, Gdańsk,
Poland, 142–155.
V.R. Pratt (1995), The Stone Gamut: A Coordinatization of Mathematics. In Proceedings of the conference Logic in Computer Science IEEE Computer Society.
H.H. Schaefer, Topological Vector Spaces. Third printing, Springer-Verlag, 1970.
Department of mathematics and statistics, McGill University
805 Sherbrooke St. W, Montreal, QC, H3P 1S4
Email: barr@barrs.org
This article may be accessed via WWW at http://www.tac.mta.ca/tac/ or by anonymous ftp at ftp://
ftp.tac.mta.ca/pub/tac/html/volumes/17/1/17-01.{dvi,ps,pdf}
THEORY AND APPLICATIONS OF CATEGORIES (ISSN 1201-561X) will disseminate articles that
significantly advance the study of categorical algebra or methods, or that make significant new contributions to mathematical science using categorical methods. The scope of the journal includes: all areas of
pure category theory, including higher dimensional categories; applications of category theory to algebra,
geometry and topology and other areas of mathematics; applications of category theory to computer
science, physics and other mathematical sciences; contributions to scientific knowledge that make use of
categorical methods.
Articles appearing in the journal have been carefully and critically refereed under the responsibility of
members of the Editorial Board. Only papers judged to be both significant and excellent are accepted
for publication.
The method of distribution of the journal is via the Internet tools WWW/ftp. The journal is archived
electronically and in printed paper format.
Subscription information. Individual subscribers receive (by e-mail) abstracts of articles as
they are published. Full text of published articles is available in .dvi, Postscript and PDF. Details will
be e-mailed to new subscribers. To subscribe, send e-mail to tac@mta.ca including a full name and
postal address. For institutional subscription, send enquiries to the Managing Editor, Robert Rosebrugh,
rrosebrugh@mta.ca.
The typesetting language of the journal is TEX, and LATEX2e is
the preferred flavour. TEX source of articles for publication should be submitted by e-mail directly to
an appropriate Editor. They are listed below. Please obtain detailed information on submission format
and style files from the journal’s WWW server at http://www.tac.mta.ca/tac/. You may also write
to tac@mta.ca to receive details by e-mail.
Information for authors.
Managing editor. Robert Rosebrugh, Mount Allison University: rrosebrugh@mta.ca
TEXnical editor. Michael Barr, McGill University: mbarr@barrs.org
Transmitting editors.
Richard Blute, Université d’ Ottawa: rblute@uottawa.ca
Lawrence Breen, Université de Paris 13: breen@math.univ-paris13.fr
Ronald Brown, University of North Wales: r.brown@bangor.ac.uk
Aurelio Carboni, Università dell Insubria: aurelio.carboni@uninsubria.it
Valeria de Paiva, Xerox Palo Alto Research Center: paiva@parc.xerox.com
Ezra Getzler, Northwestern University: getzler(at)math(dot)northwestern(dot)edu
Martin Hyland, University of Cambridge: M.Hyland@dpmms.cam.ac.uk
P. T. Johnstone, University of Cambridge: ptj@dpmms.cam.ac.uk
G. Max Kelly, University of Sydney: maxk@maths.usyd.edu.au
Anders Kock, University of Aarhus: kock@imf.au.dk
Stephen Lack, University of Western Sydney: s.lack@uws.edu.au
F. William Lawvere, State University of New York at Buffalo: wlawvere@acsu.buffalo.edu
Jean-Louis Loday, Université de Strasbourg: loday@math.u-strasbg.fr
Ieke Moerdijk, University of Utrecht: moerdijk@math.uu.nl
Susan Niefield, Union College: niefiels@union.edu
Robert Paré, Dalhousie University: pare@mathstat.dal.ca
Jiri Rosicky, Masaryk University: rosicky@math.muni.cz
Brooke Shipley, University of Illinois at Chicago: bshipley@math.uic.edu
James Stasheff, University of North Carolina: jds@math.unc.edu
Ross Street, Macquarie University: street@math.mq.edu.au
Walter Tholen, York University: tholen@mathstat.yorku.ca
Myles Tierney, Rutgers University: tierney@math.rutgers.edu
Robert F. C. Walters, University of Insubria: robert.walters@uninsubria.it
R. J. Wood, Dalhousie University: rjwood@mathstat.dal.ca
Download